OnCo
ideasIdea

A shared AI review assistant that maps one dossier to every regulator's questions

Regulators are starting to use AI to read dossiers faster. If they shared one tool, it could show where their questions overlap and where they truly disagree.

FDA deployed an internal generative AI assistant agency-wide in 2025; EMA and MHRA have pilots. Each is used in isolation. The proposal is a jointly governed assistant with an audit trail that, for a given dossier, drafts the assessment sections for each participating agency's template, flags inconsistencies between the dossier and the trial registry, and surfaces where two agencies' historical decisions on similar evidence diverged. Reviewers retain all decisions; the tool exposes divergence.

Hypothesis
Reviews using the shared assistant take at least 30% fewer reviewer-hours per oncology dossier and produce information requests to sponsors that overlap by more than 80% between agencies, up from an unmeasured but lower baseline.
Rationale
The bulk of review time is reading and cross-referencing, which large language models do well under supervision; a shared tool makes the tacit differences between agencies explicit and therefore fixable.
What would test it
Run the assistant retrospectively on 30 completed Orbis dossiers and compare its drafted assessments with the agencies' actual reports; then run prospectively in a reviewer-in-the-loop pilot.
Maturity
early clinical
Who has to act
regulator
Cost to try
Medium ($1M to $50M)
Years to first evidence
3
Bottlenecks it attacks

Connected

2top

Pages like this

not linked directly; found by shared links

bottlenecks

2
AI that is built but not validated or deployedRegulatory divergence between regions